Transcript ppt - UiT

Conceptual Overlap and the
Illusion of Semantic Emptiness
Laura A. Janda
Tore Nesset
and the CLEAR group at the
University of Tromsø
1
Who is CLEAR?
•
•
•
•
•
•
Laura A. Janda, Tore Nesset
Olga Lyashevskaya
Svetlana Sokolova
Julia Kuznetsova
Anna Baydimirova
Anastasia Makarova
CLEAR:
Cognitive Linguistics:
Empirical Approaches
to Russian
11.04.2017
2
Why use quantitative approaches in
cognitive linguistics?
•
Usage-based approaches:
–
–
–
•
Categorization:
–
–
•
Language system and language use are not separate:
Generalizations grow out of language use
Linguists must study actual language use
Not all categories have clear-cut boundaries
Gradient phenomena are acknowledged
The information revolution:
–
–
Large electronic corpora available
Tools for handling large amounts of data needed
Cognitive linguistics needs statistical methods.
11.04.2017
3
Pioneers: Collostructional analysis
• Which words fit into a construction?
• Example: NP waiting to happen
• Whether a word fits is a matter of degree
of (repulsion or attraction)
– E.g. disaster, accident are attracted to the
construction
• Stefanowitsch and Gries (2003, 2004 etc.)
developed statistical methods for the
analysis of repulsion and attraction
• Objective description of a word’s
relationship to a construction
11.04.2017
4
Behavioral profiles
• Are near synonyms really different?
• Divjak & Gries studied 1585 sentences with
9 verbs of trying in Russian
• Each sentence tagged manually for 87
variables (aspect, clause structure …)
• Each verb receives percentage for each
variable
• Each verb has a “behavioral profile” defined
by its values for the variables
• Behavioral profiles can be analyzed
statistically
• Objective description of differences and
similarities among near synonyms
11.04.2017
5
Constructional profiles
• Janda & Solovyev (2008) studied
– Nouns for sadness/happiness in Russian (near
synonyms)
– 70 constructions: (Prep) + NounCASE
• Constructional profile:
– “The distribution of relative frequencies of
constructions associated with a given word”
• Constructional profiles can be compared by
means of statistical analysis
• Objective description of syntactic similarities
and differences between near synonyms
11.04.2017
6
Grammatical profiles
• Janda & Lyashevskaya (to appear) study
token frequencies of inflected forms of
Russian verbs (nearly 6 millions)
• Verbs show remarkably different behavior
• Grammatical profile:
– “Relative frequency distribution of the inflected
forms of a word in a corpus”
• Grammatical profiles can be compared by
means of statistical analysis
• Grammatical profiles shed light on the nature
of aspectual pairs in Russian
11.04.2017
7
Radial Category
Profiling
D
D
C
A
B
A
C
B
• Subcategories have different numbers of members
(type frequencies)
• Radial Category Profile: The relative frequency
distribution of the subcategories of a radial category
• Profiles of different categories can be compared with
simple statistical methods
• Case study: Janda, Nesset & Baydimirova (in press)
11.04.2017
8
Semantic profiles
• Janda & Lyashevskaya (to appear) study
attraction and repulsion between
– Russian aspectual prefixes and
– Semantic classes of verbs (tagged in the
Russian National Corpus)
• Prefixes show remarkably different behavior
• Semantic profile of a prefix:
– Relative frequency distribution of the semantic
classes of verbs in a corpus that combine with
a prefix
• Semantic profiles can be compared by means
of statistical analysis
11.04.2017
9
Conceptual Overlap
• Is a linguistic unit ever semantically “empty”?
• If a linguistic unit, like a prefix, never appears in isolation,
it can be hard to say what its meaning is
• Though some claim that such bound morphemes are
“empty”, they may instead show conceptual overlap
• Methods for exploring meaning in situations of
conceptual overlap:
– Radial Category Profiling
– Semantic Profiling
10
Conceptual Overlap
• “Redundancy is not to be disparaged, for in
one way or another every language makes
extensive use of it” (Langacker 2008, 188)
• Conceptal overlap is found in common
collocations such as added bonus and physical
exercise
• Hypothesis: The meaning of a bound
morpheme and the lexical morphemes it
attaches to show conceptual overlap
11
Are Russian prefixes empty?
• Conventional wisdom:
– Purely aspectual prefixes are
semantically “empty”
• Our alternative Hypothesis:
– Conceptual overlap
• How can this be tested empirically?
– Radial Category and Semantic Profiling:
–Corpus data
–Statistical analysis
12
Overview
• General arguments why prefixes aren’t empty
– Number and distribution of prefixes
– Borrowings
– Prefix variation
• Case study of the raz- prefix
– Used in some types of perfectives with spatial
meaning
– Claimed to be “empty”
• Remaining prefixes and methodology
– Radial Category Profiling for “small prefixes”
– Semantic Profiling for “big” prefixes
13
RAZojtis’ ‘walk in different directions’
John Cleese in the Monty Python sketch “Ministry of silly walks”
14
Russian aspectual prefixation
RAZ-tajat’
‘melt’ pf
Specialized
perfective
Lexical prefix
tajat’ ‘melt’ ipf
vit’ ‘twist’ ipf
žeč’ ‘burn’ ipf
RAZvit’
‘develop’
pf
Natural perfective
Purely perfectivizing
prefix
Complex act
Superlexical prefix
RAZ-žeč’
‘kindle’ pf
15
Russian aspectual prefixation
We focus on
this part
Natural
perfective
Purely perf
prefix
This part has been
studied a lot
Imperfective
Specialized
perfective
Lexical prefix
Affects
argument
structure
Adverbial
meanings
Complex
act
Superlex
prefix
16
Why purely perfectivizing prefixes
aren’t empty (1)
• Assume:
– Only purpose of prefixes is to mark
perfective aspect
• How many prefixes are needed?
– Reasonable answer: ONE
• Russian has 19 relevant prefixes
(Krongauz 1998)
M.A. Krongauz
The number of prefixes suggests that they are
not pure markers of aspect.
17
Why purely perfectivizing prefixes
aren’t empty (2)
• Assume
– Prefixes are pure aspectual markers
• Prediction
– Even distribution of prefixes across base verbs
The UNeven distribution suggests that the prefixes
do different jobs.
18
Why purely perfectivizing prefixes
aren’t empty (3)
• Assume
– Prefixes are pure aspectual markers
• Prediction
– Prefixes are assigned to borrowings in
random fashion
ZA-asfal’tirovat’ COVER
• But
– Native speakers have intuitions
– Borrowings are assigned prefixes in a
consistent way.
PRO-fil’trovat’
MOVE
THROUGH
The consistent assignment of prefixes to borrowings suggests that
prefixes are not semantically empty.
19
Structure of the argument
1. Explore meaning of raz- in verbs where its meaning is
UNcontroversial:
– Specialized perfectives (lexical prefixes)
– Complex act perfectives (superlexical prefixes)
2. Compare with the use of raz- in verbs where its
meaning is controversial:
– Natural perfectives (purely aspectual prefixes)
3. The same meaning attested in (1) and (2).
4. Raz- has the same meaning in all types of perfectives.
5. There is no semantically empty raz- in Russian.
20
Meaning: A network model
• Category:
– Network of related subcategories
• Prototype:
– Central subcategory that is the best example of the
category as a whole
• Extension relations:
– Subcategories relate to the prototype via e.g.
metaphor and metonymy.
• Schema:
– Categories may have a general schema that covers all
subcategories.
21
General schema and prototype for raz• “APART”:
– Outward movement in various
directions from a point
‘To explode’ is
RAZorvat’sja
• The general schema is instantiated in a variety of
subcategories
• Prototype = “PHYSICAL APART”
– Physical object divided in pieces
22
Specialized/complex act
perfectives
9. UN-, DIS-
10. UN-, DIS(metaphor)
razgruzit’ ‘unload’
rastoptat’
‘trample’
1. PHYSICAL
APART
2. CRUSH
4. SPREAD
(metaphor)
5. SOFTEN,
DISSOLVE
raspilit’ ‘saw apart’
6. SWELL
3. SPREAD
rastvorit’
‘dissolve’
rasšifrovat’ ‘decipher’
7. EXCITE
11.
INGRESS.
raskalit’
razvolnovat’sja
razreklamirovat’
‘make redraskatat’
‘become
upset’‘roll out’
‘publicize all over’
hot’
razdut’ ‘inflate’
8. EXCITE
(metaphor) razdosadovat’
23
‘annoy’
Natural perfectives
9. UN-, DIS-
1. PHYSICAL
APART
2. CRUSH
4. SPREAD
(metaphor)
10. UN-, DIS(metaphor)
5. SOFTEN,
DISSOLVE
6. SWELL
3. SPREAD
7. EXCITE
11.
INGRESS.
8. EXCITE
(metaphor)
24
Only in
specialized
perfectives
Natural perfectives
9. UN-, DIS-
1. PHYSICAL
APART
2. CRUSH
4. SPREAD
(metaphor)
10. UN-, DIS(metaphor)
5. SOFTEN,
DISSOLVE
6. SWELL
3. SPREAD
7. EXCITE
11.
INGRESS.
Only in complex
acts
8. EXCITE
(metaphor)
25
Semantic overlap and the illusion
of emptiness
Specialized perfectives &
complex acts
APART
VERB
MEANING
RAZ- VERB STEM
• Prefix and verb have
different meanings
• The meaning of the prefix
stands out
Natural perfectives:
APART
VERB
MEANING
RAZ- VERB STEM
• Prefix and verb have
overlapping meanings
• The meaning of the prefix is
“invisible”
• An illusion of semantic
26
emptiness is created
Radial Category Profiling
• A method for comparing meanings
– Radial category for Specialized & Complex Act
Perfectives
– Radial category for Natural Perfectives
– We see that the base verbs of the Natural
Perfectives have the same range of meanings as
posited for the prefixes in Specialized & Complex
Act Perfectives
– Radial Category Profiling reveals conceptual
overlap between verbs and prefixes
27
Further use of
Radial Category Profiling
• The “small” prefixes (entire CLEAR group)
– u-, ot-, pri-, v-, raz-, vz-/voz-, vy-, iz-, pere-, and pod(over 1300 verbs analyzed)
– For all 10, the two radial categories coincide
• 3 have 100% overlap, 5 majority overlap, 3 minority
(contiguous) overlap
• Meanings not among NPs are phasal, annulment,
quantitative comparison, repetition
• Related prefixes: vy-, iz-; o-/ob-/obo28
Semantic Profiles
• The “big” prefixes: po-, s-, za-, na-, pro– Thousands of verbs and diffuse meanings make
Radial Category Profiling problematic
– Analysis of semantic tags assigned to verbs in Russian
National Corpus
• Moscow semantic school
• independent, objective measure
• focused on these tags: IMPACT, CHANGE STATE, BEHAVIOR,
SOUND&SPEECH
– 382 verbs analyzed (all existing NPs with these
prefixes, single prefix and single tag)
29
Semantic Profiles: Results
• Each prefix does have a unique semantic
profile
• Chi-square analysis shows that there are
significant differences (chi-square = 248, df =
12, p = 2.2e-16, effect size, Cramer’s V = 0.81)
• Additional calculation of Expected Values and
Fisher Test determine which semantic tags
each prefix is attracted to and repulsed from
30
Semantic Profiles
• pro– Attracted to SOUND&SPEECH (sounds that carry
through space or time)
– Neutral to IMPACT (penetration)
– Repulsed from BEHAVIOR, CHANGE STATE
• po– Attracted to CHANGE STATE, SOUND&SPEECH (increase
along a scale, duration)
– Neutral to IMPACT
– Repulsed from BEHAVIOR
31
Semantic Profiles
• za– Attracted to IMPACT, CHANGE STATE (covering, filling,
fixing)
– Repulsed from BEHAVIOR, SOUND&SPEECH
• s– Attracted to BEHAVIOR (semelfactive)
– Neutral to CHANGE STATE, SOUND&SPEECH, IMPACT
32
Semantic Profiles
• na– Attracted to IMPACT, BEHAVIOR (accumulate on a
surface, large quantity)
– Neutral to SOUND&SPEECH
– Repulsed from CHANGE STATE
33
Semantic Profiles
• Summary of results
– The meanings of the verbs with “empty” prefixes
(Natural Perfectives) as classified by their semantic tags
correspond to the meanings of the prefixes in their
“non-empty” uses as previously described by scholars
– Conceptual overlap: each verb selects the prefix that
conforms best to the verb’s meaning
34